// Use this skill to build and test code changes inside a cudf devcontainer.
Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV/Parquet I/O, nullable semantics, and multi-GPU DataFrame workloads.
Debug and fix pandas test suite failures under the cudf.pandas compatibility layer. Use when given pytest node IDs of failing pandas tests that need to be fixed for cudf.pandas compatibility.
Use this skill to review GitHub pull requests for cudf
Build and test cudf Java bindings (cudf-java) inside a cudf devcontainer. Use when the user asks to build, compile, or test Java code in the cudf repository.
| name | build-test-cudf |
| description | Use this skill to build and test code changes inside a cudf devcontainer. |
Detect cudf devcontainer usage by checking if the username is coder. If not, ignore everything else in this file and skip.
cudf packages must be built in this order as needed (each depends on the previous):
.pyx files)When rebuilding a package, ensure its dependencies are already built. Since we are in cudf devcontainer, DO NOT run the build.sh script or install any packages yourself. Instead, always use the build-*, test-*, rapids-*, clean-* scripts located at /usr/bin.
Whenever building libcudf from scratch (CMake has not been run before), run:
build-cudf-cpp -j0
For incremental builds, you can also just navigate to cpp/build/latest and run:
ninja
Both build-cudf-cpp and configure-cudf-cpp accept CMake -D options directly as arguments. See cpp/CMakeLists.txt for a full list of available CMake options
# Default option
build-cudf-cpp -j0 -DBUILD_BENCHMARKS=ON
# Multiple options can be combined
build-cudf-cpp -j0 -DBUILD_BENCHMARKS=ON -DBUILD_TESTS=OFF
Similarly, configure without building:
configure-cudf-cpp -DBUILD_BENCHMARKS=ON
build-libcudf-python
build-pylibcudf-python
build-cudf-python # cudf-python
build-cudf-polars-python # cudf-polars
Similar to build instructions, we can use clean-cudf-xxx scripts also located at /usr/bin. Use clean-cudf to clean everything.
If there are build errors, cleaning before building will usually resolve problems. If we run into a fatal CMake error while building libcudf indicating packages/version mismatch, update the environment using the following script. If the error persists, use --force flag with the script
rapids-make-${PYTHON_PACKAGE_MANAGER}-env
Make sure to run clean-cudf after running this for a fresh subsequent build.
Make sure that libcudf has been built before running any of these. Discover all Google test name binaries at cpp/build/latest/gtests and run relevant Google tests using test-cudf-xxx script located at /usr/bin.
test-cudf-cpp -j10 # all tests 10 parallel jobs
test-cudf-cpp -R <NAME>_TEST # specific test suite
cd cpp/build/latest/gtests && <NAME>_TEST --gtest_filter="<pattern>" # Run tests matching the <pattern> from the <NAME>_TEST
These instructions are applicable for running tests for pylibcudf and high-level python packages. Make sure that the package has been built before running its pytests. Pylibcudf tests can be discovered at: python/pylibcudf/tests and subfolders. Run them using:
pytest python/pylibcudf/tests/<subfolder>/test_<name>.py # specific pylibcudf test suite
cudf-python pytests can be discovered at: python/cudf/cudf/tests and subfolders. Run them using:
test-cudf-python # run all cudf-python Pytests
pytest python/cudf/cudf/tests/<subfolder>/test_<name>.py # specific cudf-python test suite
Similarly, cudf-polars pytests can be discovered at: python/cudf_polars/tests and subfolders. Run them using:
test-cudf-polars-python